Re: an autoconf expert challenge

[Ralf Wildenhues]

Hello Jim, all,

* Jim Meyering wrote on Wed, Jan 24, 2007 at 04:21:38PM CET:
> 
> However, for a full solution, we need a mechanism
> to keep such problems from being reintroduced.

Not just that.  The fact that issues are fixed in the gnulib test
doesn't mean that all issues are fixed in, say, gettext's usage of
gnulib.  I've thought a bit about letting M4 go through all possible
expansions in order to find more such issues, but that isn't feasible
with today's Autoconf code AFAICS: it depends in several places on
the fact that what we expand, we really want to be expanded.  So it
usually fails early on with some assertion guard that failed.

> Ralf, can you suggest something that would be a little more durable?
> At worst, we can just use that guaranteed-to-change-someday code as-is,
> but only for testing.

Well, you could let gnulib-tool emit it for Autoconf 2.60 and 2.61.
Not a nice prospect, I agree; but I'd definitely try it on the actual
package code as well.

Cheers,
Ralf